MCP Toolbox for Databases

MCP Toolbox for Databases is a Google-stewarded open-source Model Context Protocol server that gives LLM agents structured, governed access to real production databases. Out of the box it ships a set of generic tools (list_tables, execute_sql, semantic search) so you can point Claude, Gemini, or any MCP client at Postgres, MySQL, MongoDB, BigQuery, Snowflake, Spanner, AlloyDB, and dozens of other engines without writing glue code. Beyond exploration, it's also a framework: you declare typed, parameterised tools in YAML, and the server enforces what the agent can and can't run.

The pitch is aimed at teams who don't want their agents loose with raw SQL credentials. You get connection pooling, IAM-backed auth, OpenTelemetry traces, and dynamic config reloads, plus SDK bindings for Python, JavaScript, Go, and Java so the same tool definitions work in LangChain, LlamaIndex, Genkit, and Google's Agent Development Kit. It's fully open source under the googleapis org, free to self-host, and there's a small UI for poking at toolsets while you develop.

The trade-off is that it's infrastructure, not a SaaS — you run it yourself, configure tools.yaml, and operate the server. There's no managed tier, and on the agent side you still bring your own LLM. For shops already on Google Cloud the integrations are tightest, but the project is genuinely engine-agnostic.